---
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: image
    dtype: image
  - name: shape_type
    dtype:
      class_label:
        names:
          '0': 0
          '1': 1
          '2': 2
          '3': 3
          '4': 4
  - name: shape_name
    dtype:
      class_label:
        names:
          '0': circle
          '1': rectangle
          '2': hexagon
          '3': star
          '4': triangle
  - name: pos_x
    dtype: float32
  - name: pos_y
    dtype: float32
  - name: size
    dtype: float32
  - name: color
    dtype: float32
  - name: rotation
    dtype: float32
  - name: id
    dtype: int32
  splits:
  - name: train
    num_bytes: 953573
    num_examples: 5000
  - name: validation
    num_bytes: 191415
    num_examples: 1000
  - name: test
    num_bytes: 190454
    num_examples: 1000
  download_size: 924092
  dataset_size: 1335442
---

# Parametric Shapes Dataset
This is a synthetic dataset of images containing geometric shapes with controllable parameters generated for educational purposes at the University of Toronto.

This dataset contains 7000 synthetic images (32x32) of geometric shapes including circles, rectangles, hexagons, stars, and triangles. Each image is generated with random parameters such as position, size, color, and rotation.

## Dataset Splits

Train: 5,000 images (1,000 per shape)
Validation: 1,000 images (200 per shape)
Test: 1,000 images (200 per shape)

All splits are non-overlapping and generated with different random seeds.

## Parameters

Each image is parameterized by 6 independent factors:

| Factor | Type | Range | Description |
|--------|------|-------|-------------|
| `shape_type` | Categorical | 0-4 | Shape class (0=circle, 1=rectangle, 2=hexagon, 3=star, 4=triangle) |
| `pos_x` | Continuous | 0.2-0.8 | Horizontal position of shape center (normalized) |
| `pos_y` | Continuous | 0.2-0.8 | Vertical position of shape center (normalized) |
| `size` | Continuous | 0.1-0.35 | Size of shape relative to image (radius/width) |
| `color` | Continuous | 0-1 | Hue value (converted to RGB with high saturation) |
| `rotation` | Continuous | 0-2π | Rotation angle in radians (no effect on circles) |



## License

This dataset is released under the MIT License. Feel free to use it for educational and research purposes.